How much data do you need to run an MMM?
Feb 18, 2024
Introduction
Remember the term “statistical significance” from high school statistics? It refers to the likelihood of data results having a correlational or causal explanation, not just being a matter of chance. The larger an experiment’s sample size, generally the more statistically significant the results are.
The same logic applies to marketing mix modeling (MMM). An MMM needs enough data to draw out valid correlations between your marketing inputs and business outcomes.
Most marketing leaders using MMM understand this point, but they overestimate how much data their model needs. Just Google the question, "How much data does my MMM need?" and you'll see article after article promoting complicated statistical tools with data jargon to answer the question.
None of that is necessary. Our team has analyzed over $100M in marketing spend to determine precisely how much data you need. In this article, we'll provide simple ways to evaluate whether you have sufficient data for MMM to work for you and debunk popular myths. If you don't have enough data for MMM, we've got you covered with an alternative solution.
What amount of data does my MMM need?
We recommend at least twelve months of data on a daily aggregation, and your target metric should be in the hundreds every week. For example, you're in good territory if your business generates hundreds of orders, leads, or demo sign-ups weekly.
If your company only has one or two marketing channels, you likely don't even need one! A marketing leader working at a small business in this stage can use simpler reporting methods to gauge their campaign and channel performance.
Paramark, for example, is at this early phase where we don’t need MMM yet. We're a three-person company with two main marketing channels, social media and community referrals. Given our limited channels, our team can evaluate marketing performance by looking at statistics provided by LinkedIn and community platforms (impressions, likes, comments).
What data do I need for my MMM?
The metrics you include in your model depend on your business's sales cycle length and marketing goals. But from a general perspective, an MMM should track:
Input metrics: KPIs that reflect your marketing activity, such as the number of daily emails sent.
Output metrics: KPIs representing sales directly or indirectly, depending on your sales cycle length. A B2B brand that has a long sales cycle will need a leading output metric, like qualified traffic or leads, to gauge sales, while a B2C business with a short customer journey can measure sales directly in their MMM.
Cost: The amount you’re spending to run your marketing inputs.
This post will focus on gauging the amount of MMM data needed rather than what information should be tracked. To learn more about picking your MMM’s metrics, check out our post, “What is marketing mix modeling?”
What are the signs of not having enough data in your MMM?
The simple answer is your MMM data doesn’t meet the requirements mentioned previously:
You have less than 12 months of data.
Your marketing inputs aren’t aggregated on a daily basis.
Your target metric isn’t in the hundreds per week.
But what about in-between situations? Maybe you have weekly data rather than daily, but it spans five years. Because your reporting covers an especially long period, your weekly data is actually sufficient for your MMM.
A step that Paramark takes with customers is to evaluate the validity of the MMM by measuring the "fit" of your MMM. This is a way to understand how well the model predicts your target metric. A model with poor fit doesn’t necessarily lack data, but it’s one factor that could cause it to produce inaccurate correlations.
Our first step in calculating fit is using 90% of a client’s data in an MMM to form predictions about their target metric. From there, we compare those predictions against the remaining 10% of data—the actual target metric results—using mean absolute percentage error (MAPE). This metric represents the average deviation between predicted and actual values in your model. In other words, MAPE measures the accuracy of your model’s predictions.
Say your MMM has a MAPE of 20%. That means your actual marketing performance deviated on average by 20% from your model’s predicted results. If your MAPE is higher than 10%, take a hard look at your dataset. Your MMM likely needs more information if you don’t have at least one year of data with daily aggregation.
If your data volume is sufficient, study your model with other team members to consider alternative reasons for its poor fit. For example, a holiday season spike in sales may explain the large difference between your MMM’s predicted and actual results.
Some teams use R-squared to assess their model’s fit, but we prefer MAPE. An R-squared score doesn’t reflect a model’s predictive ability—it measures its variance—so it doesn’t tell you whether your MMM’s fit is good or bad.
Whatever method you use, it’s best to have a marketing analyst or data scientist determine your model’s fit since the process is technical. Or, consider a marketing analytics platform like Paramark that can build your MMM and evaluate its accuracy.
What do I do if I don’t have enough data for my MMM?
In this situation, set up experiments to gauge your marketing’s impact on sales. Incrementality testing seems daunting but is fast and efficient in offering powerful insights if run correctly. Use experiments to verify which marketing activities contribute most to sales rather than just finding the correlation with MMM. We also recommend adding testing data to your MMM to help it produce valid outputs.
3 myths about MMM data requirements
Many online resources about MMM claim models need extensive amounts and types of data, so they’re only suitable for the biggest enterprises.
While it’s true that MMM isn’t for every company, its data requirements aren’t so restrictive. Our team has seen MMM work for various companies of different sizes and industries. Based on our work, we don’t buy these common MMM data requirement myths.
You need five years of data
This claim isn’t universally true because it doesn’t account for data collection frequency.
Say your team tracks its target metric sales on a daily basis, and your campaigns generate hundreds of sales every week. In this case, a year’s worth of reporting is enough for a marketing mix model.
On the other hand, imagine a company that only tracks sales quarterly. Even five years of this data wouldn’t be enough to establish a valid correlation between the business’ marketing activity inputs and target metric.
TL;DR: Years alone don’t indicate whether you have sufficient data for your MMM.
I need tens of millions of dollars in marketing spend
Not quite. The validity of your MMM all boils down to your marketing data’s volume and granularity, not your spend levels.
Imagine a business with $10 million in annual revenue that generates anywhere from 200 to 600 weekly sales. The company isn’t as large as Nike or McDonald’s, but their target metric is in the hundreds every week. Their MMM will likely be reliable with a year or more of this data.
My business needs a short sales cycle
Many marketers assume MMM only works for businesses with short customer journeys, like an ecommerce brand. These companies can impact revenue relatively quickly with marketing, so they can use sales as their target output metric in their MMM.
That’s true, but it doesn’t mean businesses with long customer journeys can’t use MMM. These brands can develop valid models by measuring their marketing’s impact with leading metrics—such as qualified traffic or demo sign-ups—and a significant volume of data.
Don’t let technical jargon scare you away from MMM
It's easy to get lost in the maze of technical jargon and myths around MMM. At Paramark, we're here to demystify the modeling process. Don't let the naysayers scare you away – MMM is within reach for businesses of all sizes and industries.
Ready to demystify your MMM? Let’s build your model and assess its performance and accuracy together. Sign up for a free demo call with Paramark's team today.
Solutions